COMPREHENSIVE EXAMINATION OF DEHYDROXYLATION OF KAOLINITE, DISORDERED KAOLINITE, AND DICKITE: EXPERIMENTAL STUDIES AND DENSITY FUNCTIONAL THEORY

— Kaolinsandclaysareimportantrawmaterialsforproductionofsupplementarycementitiousmaterialsandgeopolymerprecursors through thermal activation by calcination beyond dehydroxylation (DHX). Both types of clay contain different polytypes and disordered structuresofkaolinitebutlittleisknownabouttheimpactofthelayerstackingofdioctahedral1


INTRODUCTION
Worldwide, several million tons of metakaolin are manufactured every year by calcination of kaolin for various applications, e.g. as concrete additive, supplementary cementitious material, and geopolymer precursor but also as fillers and coating for specialty paper, or as a paint extender.
The behavior of clay minerals during calcination determines optimal activation of pozzolanic reactivity or reactivity as a geopolymer precursor.Activation is achieved upon dehydroxylation of the octahedral sheet.Thereby surface reactivity is improved.Overheating results in particle agglomeration and crystallization of inactive high-temperature phases.The temperatures at which dehydroxylation and recrystallization occur are determined by the clay mineral structure (Snellings et al. 2012).
The dehydroxylation process has to be understood in detail for each clay mineral structure to tune structures of dehydroxylates (meta clay minerals) and to extend application fields.
During calcination of kaolin, its main mineral phasekaoliniteis dehydroxylated into metakaolinite (Brindley and Nakahira 1959) with O r representing residual oxygen.
In any case, a broad temperature gap between dehydroxylation and recrystallization favors the occurrence of metakaolinite and, if calcination is terminated prior to the formation of a defect, Al-Si spinel or even mullite at higher temperatures (Brown et al. 1985).The metakaolinite is highly reactive in alkaline solutions.This reactivity is explained by the transition of octahedrally coordinated Al (Al VI ) to 5-fold coordination (Al V ), established experimentally by the occurrence of a peak at ~30 ppm in 27 Al nuclear magnetic resonance (NMR) spectra (Rocha andKlinowski 1990a, 1990b).
On the other hand, the octahedral Al in a fully dehydroxylated kaolinite must end up as 4-fold coordinated (Al IV ) and doubts about the assignment of peaks at 30 ppm in 27 Al NMR spectra to Al V are well founded.The occurrence of this resonance is more likely related to the presence of Al in much distorted tetrahedral sites (MacKenzie and Smith 2002).
While disordered kaolinites dehydroxylate at lower temperatures than well ordered kaolinites, controversial results exist about the reactivity in alkaline solutions.Some studies found that dehydroxylated disordered kaolinites show a better reactivity in alkaline solutions (Kakali et al. 2001;Bich et al. 2009;Snellings et al. 2012;Hollanders et al. 2016) while other authors reported contrary findings (e.g.Tironi et al. 2014).These observations need deeper investigation to understand the relation between the dehydroxylation behavior, the structure of metakaolinite, and reactivity.
Well ordered kaolinites are very rare, and kaolinites as main constituents of kaolins show a large variability in the degree of their structural disorder.Traditionally, this variability has been evaluated by empirical indices of "crystallinity," i.e. by the Hinckley Index (Hinckley 1963) calculated from peak height of certain hkl reflections in XRD patterns, the P 0 and P 2 ratio of the OH-stretching bands of FTIR spectra, or the slope ratio of the endothermal DSC or DTA dehydroxylation peak of kaolinites (Bich et al. 2009).
Such indices cannot be related directly to the type of disorder but are used commonly to compare varieties in a qualitative manner.Drits and Tchoubar (1990) discussed several models of stacking faults, e.g.±b/3 translations, ±120°rotations, displacement of octahedral vacancies, and models containing several enantiomorphic B-(vacant) and C-(vacant) layers.The latter model has been proved to be the most probable one and the diffraction patterns of many natural kaolinites can be described successfully based on this general principle (e.g.Bookin et al. 1989;Ufer et al. 2015;Sakharov et al. 2016).Based on this approach, ideal kaolinite consists of regularly stacked B-layers and the ideal dickite polytype is characterized by regularly stacked layers with vacant B and C octahedral positions (Fig. 1) and a translation of the layers of a/3.
The dehydroxylation temperature (T DHX ) of dickite is ~100 K higher than T DHX of kaolinite (de Ligny and Navrotsky 1999) and most strikingly, dilatometry measurements revealed that kaolinite shrinks while dickite expands with dehydroxylation (Schomburg and Störr 1984).This observation also needs a more systematic investigation at the atomic level and must be considered in terms of current understanding of the ideal thermal activation of various kaolin raw materials.
Previous studies using density functional theory (DFT) by White et al. (2010aWhite et al. ( , 2010b) ) or molecular dynamics (MD) by Sperinck et al. (2011) focused on the dehydroxylation of kaolinite into metakaolinite and little attention has been given to dehydroxylation of disordered kaolinite and dickite.
The present study focused on the experimental dehydroxylation of kaolinite, disordered kaolinite, and dickite; and the differences at the atomistic level of the formation of metakaolinite from well-ordered and disordered kaolinite; and of the formation of metadickite by DFT.
The first objective of the present study was to evaluate the impact of layer stacking and disorder in dioctahedral 2:1 layer silicates on the T DHX , the number of dehydroxylation steps, and the volume changes of the minerals after DHX.The second was to elucidate the nature of apparent five-fold coordinated Al in the metamaterials.

Materials
The kaolinite samples used in the present study were KBE-1 and KGa-2.The kaolinite KBE-1 (D) was provided by Amberger Kaolinwerke Eduard Kick GmbH & Co. KG (Hirschau, Germany) and originates from a residual kaolin deposit.Such kaolins typically contain quite well ordered kaolinites.The kaolinite KGa-2 is a sedimentary kaolin from a mine site in Warren County, Georgia (USA) obtained from the Source Clays Repository of The Clay Minerals Society.KGa-2 has been examined extensively in several studies (e.g.Ufer et al. 2015;Sakharov et al. 2016) and represents a disordered variety.
The <2 μm fraction of KBE-1 was separated by sedimentation to reduce the amount of accessory minerals, especially Clays and Clay Minerals mica.The sample was called KBE1_M2.KGa-2 was not separated because the amount of mica was very small.Dickite samples were from Altenberg (Saxony, Germany) and from Kohlendorf (Poland).They were used for STA measurement.

Experimental Procedures
The kaolinite content and accessory minerals were identified and quantified by X-ray diffraction (XRD) analysis of powdered samples (diameter 20 mm, top loading).A Bruker D8 Advance A25 diffractometer (Bruker AXS GmbH, Karlsruhe, Germany) equipped with a LYNXEYE XE Detector (opening degree 2.94°a nd 192 channels).Patterns were recorded between 5 and 80°2θ with CuKα radiation with a counting time of 3 s and a step size of 0.02°2θ, an automatic slit (primary side), Soller collimator of 2.5°(primary and secondary side), and an automatic knife edge above the specimen holder.The Rietveld software AUTOQUAN (Version 2.8.0;GE Inspection Technologies GmbH, Ahrensburg, Germany) was utilized for quantitative analysis.The XRD pattern in the range 19-24°2θ CuKα was used to calculate the Hinckley Index (Hinckley 1963) which mirrors the crystallinity of the kaolinite.
The DHX behavior was studied by simultaneous thermal analysis (STA).The measurements were performed on a STA 449 C Jupiter (Netzsch-Gerätebau GmbH, Selb, Germany) equipped with a thermogravimetry/differential scanning calorimetry (TG/DSC) sample holder.The STA is connected to a quadrupole mass spectrometer (MS) 403 C Aëolos (InProcess Instruments (IPI)/NETZSCH-Gerätebau GmbH, Selb, Germany) by a heated quartz glass capillary (T = 230°C).Samples were heated at 10°C/min from 35 to 1100°C in 50 mL/min streaming dry air (SynA; 79 mass% N 2 /21 mass% O 2 ) mixed with 20 mL/min N 2 (protective gas).Pt/Rh crucibles (diameter 5 mm and height 5 mm) with a punched lid were filled with 100 mg of material.An empty Pt/Rh crucible with lid served as an inert reference sample.The mass loss during thermal reactions was determined from the thermogravimetric (TG) curve.Any mass loss after dehydration (DHD) was normalized to the dry weight after DHD.
Peak decomposition and determination of peak areas of the MS curves of evolved water (m/z = 18) of the dickites was executed by means of the PeakFit software (Jandel Scientific,  The STA was also used to prepare several thermally treated kaolinite and dickite samples.The calcination temperatures were between 200 and 900°C; and were selected to achieve full DHD, partial DHX, full DHX, and structural decomposition before recrystallization of high-temperature phases.A cooling rate of 20°C/min was applied. The Al coordination and the structural changes were studied by 27 Al magic angle spinning nuclear magnetic resonance (MAS NMR) spectroscopy. 27Al MAS NMR measurements were performed at room temperature on a Bruker Avance spectrometer 400 ultrashield with a magnetic field of 9.4 T. The 27 Al MAS NMR spectra were obtained at a frequency of 104.28 MHz.The MAS NMR measurements were done using a 4 mm zirconium rotor at a spinning speed of 12 kHz in a dry nitrogen atmosphere.The chemical shift (δ) for 27 Al was referenced to Al(NO 3 ) 3 •9 H 2 O (solid, δ = 0 ppm).
Particle morphologies of the raw and heated samples were examined using a Philips XL 30 FEG environmental scanning electron microscope (ESEM; FEI Europe, Eindhoven, The Netherlands).Without further pre-treatment small amounts of sample powder were glued onto aluminum SEMholders using conductive tape (Leit-C, Plano GmbH, Wetzlar, Germany).To improve the image quality, the samples were sputtered with a thin conductive layer (5 nm Au/Pd 80/20) and were investigated using an acceleration voltage of 15 kV.

Computational Methods
In the present study, Density Functional Theory (DFT) modeling was employed for better understanding of the process of formation of metakaolinite, metadiskaolinite, and metadickite during dehydroxylation; and of their resulting properties as defined in the Vienna ab initio simulation package (VASP) (Kresse and Hafner 1993;Kresse and Furthmuller 1996) for the calculation of the electronic structure (Giraudo et al. 2015).The electron-ion interaction was described within the projector-augmented wave (PAW) scheme (Kresse and Joubert 1999).The electronic wave functions were enlarged into plane waves up to a kinetic energy cutoff of 500 eV.The Perdew-Burke-Ernzerhof (PBE) functional was used to characterize electron exchange and correlation energy within the Generalized Gradient Approximation (GGA) (Perdew et al. 1996).The optimization of atomic coordinates was implemented via a conjugate gradient, which uses the total energy and the Hellmann-Feynman forces on the atoms.The structures were presumed to be entirely relaxed when the forces on the ions were <0.01 eV/Å.The open-source XcrySDen software has

X-ray Diffraction Analysis
Measurements by XRD revealed that both KBE1_M2 and KGa-2 contained >90% kaolinite and small amounts of accessory minerals (Table 1).The Hinckley Index of KBE1_M2 was 1.63, characteristic of a well-ordered kaolinite (Komarneni  Plançon et al. 1989).The small Hinckley Index of KGa-2 of 0.5 was in accordance with published values of 0.24-0.41(Metz and Ganor 2001;Du et al. 2010;Ndlovu et al. 2015) and confirmed a high degree of structural disorder.
KBE1_M2 consists of 46-47 mass% ordered kaolinite and 50-51 mass% disordered kaolinite.The disordered kaolinite is characterized by 93% BB/7% BC stacking sequences.88% of BB sequences have no additional ~b/3 stacking errors and, thus, the kaolinite is low b-axis error-ordered.KGa-2 is characterized by 86% BB stacking sequences with a strong tendency for additional ~b/3 translations (Ufer et al. 2015) and a low abundance of BC stacking.Thus, KGa-2 is intermediate disordered.
Both dickites are also quite pure and well ordered.The dickite from Altenberg (dickite AB) actually consisted of 100% dickite while the dickite from Kohlendorf (dickite KD) contained 4% accessory minerals.

Simultaneous Thermal Analysis
Simultaneous thermal analysis of the two natural, air-dry kaolinites (Fig. 3) showed one endothermic peak in the region between 400 and 800°C, which was associated with a maximum in the MS curve of evolved water (m/z = 18) at 562°C (KBE1_M2) and at 546°C (KGa-2) due to the dehydroxylation (DHX) of kaolinite.The normalized mass loss of 13.25 mass% and 13.24 mass% for KBE1_M2 and KGa-2, respectively, reflected perfectly the kaolinite content in both materials of ~95% determined by XRD.The DHX of KGa-2 occurred at a slightly lower temperature than did DHX of the KBE1_M2 (Figs 3 and 4).
At temperatures >950°C, two further peaks were visible in the DSC curve of both kaolinites.Neither peak is associated with any gas release and can be assigned to the first and second recrystallization of kaolinite into a spinel and mullite (Mackenzie et al. 1985).
The DHX of both dickites occurred at between 500 and 900°C, in two steps.A two-step DHX of dickites was also observed in previous studies (Frost and Vassallo 1996;Franco and Ruiz Cruz 2006).The first DHX of dickite KD is very broad while dickite AB shows a sharp peak at ~650°C.The maximum of the second endothermal peak and the second Clays and Clay Minerals peak in the MS curve of evolved water (m/z = 18) was at ~679°C (Fig. 3).The degree of DHX of dickite KD was greater, at temperatures between 400 and 650°C, before full DHX is reached for both dickites (Fig. 4).The normalized mass loss of 13.95% (dickite AB) and 13.43% (dickite KD) corresponds to the dickite content determined by XRD.
The ratio of peak areas with a peak temperature below and above 665°C was ~40:60 for both dickites (Fig. 5).

NMR Measurements
The 27 Al MAS NMR spectra (Fig. 6) of the two unheated kaolinite samples showed one main resonance signal at ~0 ppm, which revealed octahedrally coordinated Al (Al VI ).Both show an additional minor resonance signal at ~75 ppm, which indicated tetrahedrally (Al IV ) coordinated Al, confirming the presence of dioctahedral mica with Al substitutions for Si in the tetrahedral sheet.
The 27 Al MAS NMR spectra of KBE1_M2 showed an additional signal between 20 and 40 ppm.Broad 27 Al signals at ~20 to 40 ppm were observed for poorly crystalline aluminosilicates, X-ray amorphous aluminosilicate glasses, or gels (MacKenzie 2000); or for minerals with 5-fold coordinated Al (Al V ) such as andalusite (Lippmaa et al. 1986), grandidierite (Smith and Steuernagel 1992), or augelite (Bleam et al. 1989).Neither amorphous material nor the above-mentioned minerals could be detected in KBE1_M2.
With increasing temperature, the overall intensity of the 27 Al MAS NMR signals of both samples decreased drastically (Fig. 7).This indicates that only a small part of the remaining short-range order is reflected, as metakaolinite is characterized by a complex amorphous structure (Brindley and Nakahira 1959;Bellotto et al. 1995).
The recrystallization products of kaolinite are spinel and mullite.Calculations based on the structural formulaa of spinel (Verwey 1935;Low and McPherson 1988;Rozita et al. 2010) and mullite (Angel et al. 1991;Lee et al. 1997;Schneider et al. 2008) showed that the amount of Al IV ranges between 25-37.5% and 56-58%, respectively (Fig. 8).Accordingly, the Al VI is in the range 62.5-75% for spinel and 42-44% for mullite (Fig. 8).The measured Al VI and Al IV ratios match the ratio in the defect spinel/inverse spinel at ~1100°C and the ratio in the mullite at ~1400°C.
Both samples showed a decreasing Al VI signal and an increasing Al IV signal as well as an increasing signal between 28 and 40 ppm with progressive DHX (Figs. 7 and 9).Thereby, the chemical shift of 28-40 ppm for the new signal is greater than published data of 20-35 ppm (Komarneni et al. 1985b;Rocha et al. 1990;He et al. 2003;Fabbri et al. 2013) and is closer to common chemical shifts of Al IV (50-80 ppm) in clay minerals.With incipient recrystallization after DHX the Al VI signal increased again and the signal between 28-40 ppm decreased (Fig. 7).The 27 Al MAS NMR spectra of both kaolinites at 1100°C showed one signal for Al VI and one for Al IV .

First-Principles Calculations
The transformation of kaolinite, disordered kaolinite, and dickite into metakaolinite, metadiskaolinite, and metadickite via several steps through dehydroxylation was investigated using DFT.First, the thermodynamic grand canonical potential was calculated for the identification of stable phases for the desorption process of water.The desorption energy alone does not allow conclusions about the stability of a specific structure, however.Rather, one has to take into account the chemical potentials μ(A i ) of the surface constituents, A i , in order to compare energetically the interfaces with different stoichiometries.The ground state of the surface is determined by the minimum of the thermodynamic grand canonical potential Ω: where F = E -TS is the surface free energy.Here, F is approximated by the total surface energy, E, assuming similar entropy contributions, S, for different adsorption configurations.In fact, the differences in vibrational free energy and electronic entropy are typically several orders of magnitude smaller than adsorption energies resulting from chemical-bond formation as found in the present case.The last term on the right-hand side accounts for the energy changes due to a possible surface charge, q, which creates a dependence on the chemical potential of electrons given here by the Fermi The transformation of kaolinite into the Al 12 Si 12 O 39 (OH) 6 phase occurred at the third step of water removal when the chemical potential of water was -1.93 eV (Fig. 11 left).The final product was a metakaolinite at a water chemical potential of -2.67 eV.A phase which had the chemical formula of Al 12 Si 12 O 39 (OH) 6 was produced.Other phases are unstable and do not exist thermodynamically.The phase transformation from disordered kaolinite into Al 12 Si 12 O 33 (OH) 18 happens at the first step of water removal and when the chemical potential of water is -2.03 eV (Fig. 12 left).The final product was metadiskaolinite when the chemical potential of water was -2.319 eV.The intersection of the three lines of purple, red, and green (Fig. 13 left) for the phase transformation from dickite into metadickite is not very clear but was much more evident  (OH) 24 took place at the second step of water removal at a water chemical potential of -2.717 eV (Fig. 14).The final product was metadickite at the chemical potential of water of -2.82 eV.
The resulting phase diagrams as a function of partial pressure of water and temperature may be compared with experimental data.Dashed lines (Figs 11,12,and 13) indicate an excellent example for the partial pressure of H 2 O equal to 10 Pa vs. temperature.The chemical potential can be related directly to experimental conditions.The pressure-and temperature-dependent deviation, Δμ(H 2 O), from the zerotemperature value obtained from the DFT calculations can be estimated within the approximation of a polyatomic ideal gas by where k B is the Boltzmann constant, p is the pressure, T is the temperature, and λ is the de Broglie thermal wavelength of the water molecule where m represents the molecular mass and ℏ is Planck's constant divided by 2. and The first stable phase of Al 12 Si 12 O 39 (OH) 6 was formed at the third step at 425°C as a result of releasing nine hydroxyl Water molecules may form when hydroxyl groups are close enough to each other upon heating.During DHX the coordination of aluminum atoms changed.The first stable phase of Al 12 Si 12 O 39 (OH) 6 (corresponding to a degree of DHX of 75%) from kaolinite is characterized by the ratio of 1:1:0 for Al IV :Al V :Al VI (Fig. 15, middle).In contrast, the ratio is 0:1:1 for Al IV :Al V :Al VI of the first stable dehydroxylated phase of Al 24 Si 24 O 72 (OH) 24 (corresponding to a degree of DHX of 50%), observed for dickite (Fig. 17,middle).
In contrast to Mackenzie et al. (1985) and White et al. (2010a), no residual OH groups were necessary to maintain the layering of the metamaterials of any of the three 1:1 layer polytypes studied (Figs. 15,16,and 17).The SEM images of the metamaterials confirmed the remaining platelet-like morphology (Figs. 15,16,and 17 left).
The DHX of kaolinite into metakaolinite, disordered kaolinite into metadiskaolinite, and dickite into metadickite must be implemented only at the stationary point by use of the Birch-Murnaghan diagram (Bleam et al. 1989).Hence, expansion or contraction of the simulation super cells in order to compute the optimized structure at the stationary point is required.The unit cells of metakaolinite and metadiskaolinite expanded after complete dehydroxylation at the stationary points (Fig. 18).The most striking finding was that the super cells of metakaolinite and metadiskaolinite at the stationary points expanded by roughly 4% and 8% in comparison with kaolinite and disordered kaolinite, respectively.In contrast, no significant changing in the super-cell volume after complete dehydroxylation of dickite was observed.
The volume of kaolinite decreased after DHX by ~0.8 to 2%, determined by dilatometer measurements of kaolins, and the volume of dickite increased after DHX by ~6.5 to 7% (Schomburg and Störr 1984).While the calculated values from the Birch-Murnaghan equation of state predict the changes of the material at the atomic level, a direct correlation with measurements performed by optical dilatometer might by difficult.One of the most important differences in polytype structures concerning the material in the present study is the differing orientation of OH groups.From a thermodynamic point of view, the hydrogen bonding of dickite is more stable than the zigzag hydrogen bonding of kaolinite between layers.This finally affects the dehydroxylation temperatures and volume changes of the super cells.T DHX,disKao < T DHX,Kao < T DHX,Dic was found experimentally and confirmed results from previous studies (e.g.Stoch andWacławska 1981a, 1981b).
H-bonding plays a considerable role in binding layers together.To discover the strongest clay mineral, the total Hbonding energy between two immediate layers was summed and divided by the area influenced.The bond length of OH determines the calculation of the H-bonding energy (Jones et al. 2006).From DFT calculations, dickite is considered to be the strongest material with an average energy of -0.032 eV/Å 2 , which is in accordance with experiments.Kaolinite and disordered kaolinite were specified as the second and third strongest materials after kaolinite with average energies of -0.031 eV/Å 2 and -0.02 eV/Å 2 , respectively.
Furthermore, two different energies (-0.03 eV/Å 2 and -0.032 eV/Å 2 ) for dickite were found, correlating to the B-C-B-C stacking sequence of layers.In contrast to that, kaolinite and metakaolinite contain only a simple B-B-B sequence of layers, which leads in the end to only one sticking energy for each material.The number of sticking energies correlated with the number of peaks that were found in the STA experiment instead of with stepwise release of inner and outer hydroxyl groups of the octahedral sheet (Frost and Vassallo 1996).
The T DHX of disordered KGa-2 is less than that of wellordered KBE1_M2.The disorder in KGa-2 is caused by b/3 translation.For dickites, no structural models are available to describe disorder by stacking faults.In any case, the two different vacancies in the dioctahedral 1:1 layers cause a twostep dehydroxylation and the a/3 translation of the 1:1 layers results in a higher dehydroxylation temperature.
The deviation for the results between experimental and simulation studies is negligible.Two main reasons are elucidated briefly here.First, results from the computational study were limited to the thermodynamic point of view.Some of the processes might be kinetically driven or kinetically hindered.Both would mean that a thermodynamic sequence of structures is not representative.Second, as White et al. (2010b) reported, more than 3×10 84 possible transformation paths exist to obtain metakaolinite from kaolinite.Thus, the results here demonstrated a possible DHX path correctly, as calculation of all possible paths is impossible.

SUMMARY AND CONCLUSIONS
Calcination of kaolin as an activation technique to enhance its reactivity with alkaline solutions is an important industrial process.Kaolin commonly contains various polytypes of dioctahedral 1:1 layer silicates that dehydroxylate during calcination.T DHX,disKao < T DHX,Kao < T DHX,Dic was confirmed ex-perimentally and by DFT calculations.Dehydroxylation of dickite occurred, thereby, in two steps and the measured amount of evolved water corresponded to 40% of the constituent hydroxyl groups during the first step of dehydroxylation.The calculated degree of DHX for the thermodynamically stable, partially dehydroxylated intermediate of dickite was 50%.Thermodynamically stable, partially dehydroxylated intermediates of kaolinite and dickite contain Al V , while metakaolinite and metadickite contain only Al IV .The observed QCC > 4 MHz confirmed that the chemical shift at ~20-40 ppm observed for metakaolinite after full DHX corresponds to a strongly distorted coordination shell of Al IV instead of Al V .
The experimentally observed decreasing volume of kaolinite and increasing volume of dickite could not be confirmed by means of first-principles calculations, which indicates a kinetic influence on the DHX and removal of the evolved water.
The results of the present study have a significant impact on industry because they depict the material which consumes the smallest amount of energy for activation.In the future, the chemical reactivity must also be supported by further theoretical calculations incorporating the electronic structure of the material.

Fig. 11 .
Fig. 11.Calculated phase diagram of kaolinite-metakaolinite transformation as a function of the water chemical potential (left) and p,T-phase diagram (right)

Fig. 13 .
Fig. 13.Calculated phase diagram of the dickite-metadickite transformation as a function of the water chemical potential (left) and p,T phase diagram (right)